Claim - New 12 CPU Core M1 SoC for iMac Pro

[deconstruct60]

Easy. Go from 8P cores to 4P cores. 4P + 2E would be 6 Cores. Then 2 * 6 = 12 ( 8P and 4E). Done.

Fair enough, but does it make sense in the context of M1 designs? This would be a radical departure and more suggestive of a multi-chip-solution (where each chip/tile consist of 4P+2E cores).

What do you think the "2 * 6 = 12" is? Two chiplets/tiles to get to 12. It is entirely consistent with Jade2C M1 design to solve the problem. This is not physically taking the whole 8P dual cluster out. It is just using what they likely already have. Turn four cores off and have four left. Zero requirement to remove them.

The pieces are likely already there for this. The more unlikely path is that Apple is running off and doing different dies than what they originally set out to use years ago.

Again, fair enough, but if they do this kind of fundamental redesign why even stick with Firestorm? Would it still be M1?

The re-design wasn't required except in the "bottom half" of the M1 Max which is a designed extension to the "base" anyway.

Like the M1 Pro and M1 Max ( Jade-chop and Jade ) likely were originally planned years ago to launch mid-2021 this easily could have launched in a more prefect timeline late 2021 or early 2022.

Perhaps Jade4C never sees light of day , but Jade2C is probably necessary for the "pro ish" desktop Apple is likely lining up to claim covered part of the Mac Pro territory so can claim they are transition complete. Even if it is just a cube (with few or no slots ). Since that is such a low volume product it would make little sense to hold it back from another desktop product with the screen built in.

An "iMac Pro" with just a "M1 Pro" really isn't what a iMac Pro 2017 was. It is a better marketing "match up" to match "Pro" of iMac to "Pro" of the entry SoC. That is an "uplift" for the classic iMac 27" 'regular' users. However, for the MP 2013 + iMac Pro 2017 users is not a full replacement. Pragmatically, neither is one with a "M1 Max". It is closer, but it is still a backslide from the upper "half" of the original iMac Pro line up.

 

[deconstruct60]

Sure but I mentioned it would be interesting to increase GPU while keeping CPU basically constant (or maybe slightly worse depending on the die interconnect properties), but it still seems odd to go 8+4 on the CPU to me. I dunno maybe.

If 4P+4E works why wouldn't 8P+4E . kept the 4E constant (on count) and actually increased bandwidth in some contexts ( each die mainly "pulling" on direct attached local RAM ). Works for the exact same reason why 4E on M1 works. Have 'saved' die space so that can allocate to GPU + NPU + Video etc. ( more specialized and more perf/watt effective at certain computations ). Not particularly blocking those with thermal leaks either from stuff probably not using on a "big core" but could be "lit up" by basic functions on a P core. .

Is Apple beating Intel Gen 12 ( alder lake) at multiple thread benchmarks at the top end? No. In part because Intel threw 8 'E-ish' cores at it. For a number of embarrassing parallel workloads don't need "big cores". That is way GPGPU computation typically beats CPU cores at Perf/Watt for a substantive number of workloads.

The pro to max made sense to me. This, I’m not as certain that Apple would do. Maybe though.

Pragmatically, the Pro is really only a cut down Max. The code names were Jade-chop and Jade respectively. So the notion that started with "Pro" (Jade-chop) and then moved to Max ( Jade) would be quite suspect. Probably done the other way around.

So the 2 cores on the Pro (Jade-chop) are only there because there are two on the Jade. The two don't "double" the consistency there.

M1 Max ( jade) is also related to Jade2C ( maybe M1 Max Duo) . If there is going to be a derivation done to enable dual chiplets then the 2E becomes a 4E on the completed package. Right back to the same baseline 4E had on the M1. Again motivated on basically the exactly same set of reasons why it makes sense on the M1.

Jade4C also makes more sense at 8E in the context of 32P and 128 GPU . That 8E allows you to get to 128GPU (presuming the overall die size is a constraint. Fixed budget... what do you spend it on.). If there is no way to get to any non Apple GPU then need to "max out" the 128 GPU if really after super high, embarrassingly parallel workloads at minimal Perf/Watt.

Part of the issue is reusing the same baseline design across multiple products to save costs. Maximize P core count is not the objective of Apple's designs. There was a huge thread here about how the A15 was going to go heavy with "super power" E cores and greatly hot rodded P cores... and Apple threw most of the additional transistor budget to ........ GPU cores and ProRes en/decoder. Apple isn't myopically focused on P cores. The folks who are will be "confused" and "dazed" at the E core count.

But even then it feels … unsatisfying as a configuration.

dropping 2 P cores from each cluster probably gives Apple some headroom on thermals (better spacing of "hot" cores) and no bandwidth hit. The disconnect on the "12 core iMac" is probably in that 12 was the new max for this new SoC offering as opposed to "iMac Pro only uses subset of Jade2C" packages . It makes lots of sense not to create something that is just useful for the iMac Pro. It also makes sense to scale down the "M1 Max Duo" if Apple put a new lower thermal cap on the "iMac Pro" enclosure.



When M2 , M3 , M4 arrive I suspect won't see huge jump in P core ( or E core ) count over an extended period of time. GPU counts may go up. New and bulked up more computational specific function units so up. But in my opinion the "P core count war with Intel/AMD/Ampere/etc" folks are going to be deeply disappointed over the long term. Apple's number priority on Perf/Watt means P core count chasing really can't be the primary objective.

 

[crazy dave]

If 4P+4E works why wouldn't 8P+4E . kept the 4E constant (on count) and actually increased bandwidth in some contexts ( each die mainly "pulling" on direct attached local RAM ). Works for the exact same reason why 4E on M1 works. Have 'saved' die space so that can allocate to GPU + NPU + Video etc. ( more specialized and more perf/watt effective at certain computations ). Not particularly blocking those with thermal leaks either from stuff probably not using on a "big core" but could be "lit up" by basic functions on a P core. .

Is Apple beating Intel Gen 12 ( alder lake) at multiple thread benchmarks at the top end? No. In part because Intel threw 8 'E-ish' cores at it. For a number of embarrassing parallel workloads don't need "big cores". That is way GPGPU computation typically beats CPU cores at Perf/Watt for a substantive number of workloads.



Pragmatically, the Pro is really only a cut down Max. The code names were Jade-chop and Jade respectively. So the notion that started with "Pro" (Jade-chop) and then moved to Max ( Jade) would be quite suspect. Probably done the other way around.

So the 2 cores on the Pro (Jade-chop) are only there because there are two on the Jade. The two don't "double" the consistency there.

M1 Max ( jade) is also related to Jade2C ( maybe M1 Max Duo) . If there is going to be a derivation done to enable dual chiplets then the 2E becomes a 4E on the completed package. Right back to the same baseline 4E had on the M1. Again motivated on basically the exactly same set of reasons why it makes sense on the M1.

Jade4C also makes more sense at 8E in the context of 32P and 128 GPU . That 8E allows you to get to 128GPU (presuming the overall die size is a constraint. Fixed budget... what do you spend it on.). If there is no way to get to any non Apple GPU then need to "max out" the 128 GPU if really after super high, embarrassingly parallel workloads at minimal Perf/Watt.

Part of the issue is reusing the same baseline design across multiple products to save costs. Maximize P core count is not the objective of Apple's designs. There was a huge thread here about how the A15 was going to go heavy with "super power" E cores and greatly hot rodded P cores... and Apple threw most of the additional transistor budget to ........ GPU cores and ProRes en/decoder. Apple isn't myopically focused on P cores. The folks who are will be "confused" and "dazed" at the E core count.




dropping 2 P cores from each cluster probably gives Apple some headroom on thermals (better spacing of "hot" cores) and no bandwidth hit. The disconnect on the "12 core iMac" is probably in that 12 was the new max for this new SoC offering as opposed to "iMac Pro only uses subset of Jade2C" packages . It makes lots of sense not to create something that is just useful for the iMac Pro. It also makes sense to scale down the "M1 Max Duo" if Apple put a new lower thermal cap on the "iMac Pro" enclosure.



When M2 , M3 , M4 arrive I suspect won't see huge jump in P core ( or E core ) count over an extended period of time. GPU counts may go up. New and bulked up more computational specific function units so up. But in my opinion the "P core count war with Intel/AMD/Ampere/etc" folks are going to be deeply disappointed over the long term. Apple's number priority on Perf/Watt means P core count chasing really can't be the primary objective.

The difference is that they are targeting very different markets: the 4P+4E are targeting the ultra efficiency mobile/SFF/quiet market, whereas this is being placed at the top of the highest computational stack (current top anyway). It's not that it wouldn't "work", it's that it wouldn't necessarily be a good fit for the market it is targeting. I agree that Apple could do it, it'd probably be okay (actually I'd like it), but it would be underwhelming (for the market they've been targeting).

My comment on the relationship between the Pro and Max was not about which was designed first, but rather that the relationship in the product stack makes sense between the two. The Pro's configuration is an okay GPU paired with a great CPU. Going up the product stack, the Max moves those to parity. It balances the two. (Yes the design was the other way around, but that doesn't change my point) Again, effectively maintaining the the same CPU with a more powerful GPU would "work" and if Apple were building a gaming rig, then I'd even say, absolutely they'd do that. But they're not. Apple's product stack is a productivity focused stack which tends towards more balanced builds - so far with emphasis on maintaining the CPU power on cutdown models. Could Apple upend this and offer a cut down config that goes all in on the GPU? Yes. But it would go against their current pattern. Truthfully I'd love a configuration that really went hog wild on Apple's GPU. I think that'd be great! But *so far* Apple hasn't shown an interest in it and I don't think it's a good fit for how the iMac Pro would be positioned. Yes there are pros who emphasize the GPU more than the CPU but those often still have high CPU demands and don't mind paying for more power in that department to get access to a better GPU.

Your stated benefits don't just apply to an 8+4 config - only to a cut down multi-die configuration vs a new single die. Which as you'll recall from the other thread (and this one), I very much agree with. I believe Apple will try to reuse as much as they can - again if you recall from the other thread, one of my rationals for why the Mac Pro would start off cheaper was that the base of the Mac Pro would start with a cut down multi-die configuration. However, a 10+2 configuration would also allow Apple to reuse the dies in a dual M1 Max config while still providing a CPU uplift. So all your benefits apply to it as well. There are all sorts of binning combinations Apple could do to achieve it and the bandwidth of two additional P cores is fine. The CPU's can't get anywhere close to maxing out the 400GB/s anyway, even with more GPU cores they'd still be okay. Thermally, tt would still be a cut down configuration and would still be far cooler than a full duo. Each P-core is only a few watts after all - even counting all the supporting infrastructure. It makes more sense to balance a 50% increase in GPU performance with a 25% increase in CPU performance.

I floated the possibility of an 8+4 configuration myself. In some ways, it makes sense. But if this rumor, that it's twelve total CPU cores, holds any water at all (and it may not!) then regardless of whether it is a cutdown multi-die or a brand new single die, then to me it makes more sense to me to increase the P-cores than the E-cores going up the stack with what Apple is doing so far. And I should note that @cmaier believes people are overstating the benefits of reusing dies and that a monolithic die might still make more sense economically here. If it is a new monolithic die with 12 cores, I can't see it being 8+4 - it could even be 12+0. I think 10+2 is the most likely.

I fully agree that future M-versions of Apple's chips, especially the high efficiency ones, will grow GPU cores faster than CPU cores. What I'd really like to see is that when multi-die solutions start trickling down the product stack (maybe around M3?), I hope that we'll get more mix-and-match CPU and GPU performance BTO options. Still tricky for Apple logistically, but it would easier to handle more mix and match options than monolithic dies! I know not hugely likely. But I'd like to see it. Also of interest is if Apple really is designing a game console what it will be.

 

[iPadified]

People seem to forget all other sub functions on M1 like hardware video encoders and ML that lessens the dependence on the GPU to some extent. Taking this into account, duo chip solution seem far better in order to scale all functions on M1/Pro/Max and not just the GPU and CPU.

 

[senttoschool]

die yields have little to do with this. Binning down working cores allows Apple to have a taller pricing ladder to more expensive ( and higher profit percentage) packages with fully working cores.

"Binning" isn't purely filled with defects. For high yields can use fully functional cores and just turn cores off. If charging full die processing recovery costs in the price of the "lowest config" package then still making a profit.

As the wafer size has gotten larger there are just more dies coming off of a wafer. For a mid-side die like the Max then even more so.

If you look at the BTO for the MBP 14/16" models can see that Apple is charging about $100/core to walk up a +2 ladder. These are not super slim margin SoCs. Apple is fully using the performance value of the SoC to slap a substantive mark up on these chips. Apple "suffering" under labor of paying for defects.... errrr No. These are priced to make more than healthy margins. The binning here is driven here far more by market segmentation usage than recovering defect overhead.

Die yields have everything to do with this. Yes, chipmakers can and do bin down based on competition and price. But why not sell an 18-core at a higher price instead? Binning down to 12 cores from a total of 20? Why? Why not 18 cores? Why not 16? Why not 14? Why bin down 40% of the CPU cores when 5nm yields are so good?

Unless Apple plans to introduce 20, 18, 16, and 14 core versions, it makes no sense to deliberately disable 40% of good cores.

 

[theluggage]

Unless Apple plans to introduce 20, 18, 16, and 14 core versions, it makes no sense to deliberately disable 40% of good cores.

Compared to making a whole new die with 12 cores just for one model? (If this rumour is true - if not this is all moot anyway).

Otherwise, the 20, 18 and 16 core options sound quite likely.

Nobody knows what sort of real profit margin Apple is making on M1 chips - it's hard to define anyway when the 'cost' of a M1 die is part of some notional internal market within Apple - but I draw your attention to Apple's standard $200 charge for adding 8GB of RAM - even on machines with bog-standard DDR4 SODIMMS ($83 retail, 1-off for 2x8GB sticks, not counting the value of the 2x4GB it replaces, including Crucial's profit margin - and who knows what sort of quantity discounts a behemoth like Apple can get - they are certainly not paying $80 a pop!) - yet if you look at machines with soldered-in LPDDR or M1 machines with on-package RAM then it's the same $200/8G across the board. Also, wasn't it a coincidence that the M1 MBA, the 13" M1 MBP and the 24" iMac all came out at exactly the same retail price as the Intel systems they replaced, despite being completely different internally? The pricing of premium products like Macs is not the result of adding up the BOM and direct manufacturing/distribution costs and adding a reasonable-sounding mark-up. It is 90% marketing strategy, what the market will bear and how much you can skimp on the base model spec so you can clean-up on upselling and upgrades.

If Apple releases a baseline 12 core "M1 Duo" at an attractive entry price to get punters "in the door" and they can then upsell people to 16, 18 or 20 cores at $$$ a step, the cost of "wasting" cores in the 16-core model is irrelevant.

The 5k iMac range currently runs from $1800 through $4400 (and beyond, but I'll stop at 64GB/1TB as a 'reasonable' spec for that CPU and GPU - the cost of maxing out RAM and SSD beyond that will probably be identical with a M1 Duo system) and of course the iMac Pro started at $5k and went steeply up from there, so I don't think Apple would be able to replace that whole range with *just* the 12-core option.

NB, re some of the original tweets: people keep on talking as if there's some objective significance to whether this is called the "iMac Pro" or not. That's 100% marketing. Going forward, it would make most sense if "Pro" written on a Mac meant it had had the M1/M2/M3 Pro or better processor (currently, only the 13" MBP breaks that rule, and it wouldn't be surprising if that model gets obsoleted by the all-new Air replacement) - naming-wise, a ~$2000 iMac Pro would make perfect sense (and, like the old range, would mean that the 'pro' iMac starting price was the same as the MacBook Pro).

The selling point of the former "iMac Pro" - which justified the price tag - was that it was Xeon/ECC based and featured 'workstation-class' (i.e. optimised for pro apps and GPU-based computing) GPUs. That distinction doesn't exist with Apple Silicon (unless they add ECC to the higher-end versions - AFAIK LPDDR5+ECC is a thing) but the obvious 'alternative' way of distinguishing the $5K machine to feature the Jade 4C (or the 2C if that doesn't turn up in a cheaper machine). I think the name "M1 Ultra" has been rumoured for those - but we know that "M1 Pro" has already been used for a far lesser CPU, so we'd be talking about an "iMac Ultra" or suchlike.

...but then, if the old iMac Pro had sold hand-over-fist then they'd have updated it, not abandoned it, so I don't see the incentive to directly replace it.

 

[LonestarOne]

...but then, if the old iMac Pro had sold hand-over-fist then they'd have updated it, not abandoned it, so I don't see the incentive to directly replace it.

The Mac Pro has not been a big seller for many years now. It’s become a high-end niche machine. The iMac Pro, probably even more so. People who need that level of power want to be able to choose their displays, hard drives, video cards, etc. They don’t want to be tied into the limitations of an all-in-one, and they don’t care about the convenience of setup.

 

[anthonymoody]

The selling point of the former "iMac Pro" - which justified the price tag - was that it was Xeon/ECC based and featured 'workstation-class' (i.e. optimised for pro apps and GPU-based computing) GPUs. That distinction doesn't exist with Apple Silicon (unless they add ECC to the higher-end versions - AFAIK LPDDR5+ECC is a thing) but the obvious 'alternative' way of distinguishing the $5K machine to feature the Jade 4C (or the 2C if that doesn't turn up in a cheaper machine). I think the name "M1 Ultra" has been rumoured for those - but we know that "M1 Pro" has already been used for a far lesser CPU, so we'd be talking about an "iMac Ultra" or suchlike.

...but then, if the old iMac Pro had sold hand-over-fist then they'd have updated it, not abandoned it, so I don't see the incentive to directly replace it.

Whether branded iMac or iMac Pro (or both?) I do think that if it's "only" the same configs as the MBP 14 and 16 Pro/Max chips, then it definitely needs to have a lot of other special sauce e.g. 128GB RAM, 30-32" 6k screen, promotion, etc.

I still think that Jade2C will wind up as a pricey top of the line iMac BTO option, and 4C will be reserved for an iMac Pro and Mac Pro.

 

[CWallace]

...but then, if the old iMac Pro had sold hand-over-fist then they'd have updated it, not abandoned it, so I don't see the incentive to directly replace it.

The Mac Pro has not been a big seller for many years now. It’s become a high-end niche machine. The iMac Pro, probably even more so. People who need that level of power want to be able to choose their displays, hard drives, video cards, etc. They don’t want to be tied into the limitations of an all-in-one, and they don’t care about the convenience of setup.

Most people I know who purchased the iMac Pro were software developers, photo editors and podcasters who purchased it for the power, the quiet and because it had a 5K display. And for them, it was more than powerful enough at those tasks to last them to this day, so they had no incentive to buy an updated model even if had been made available last year.

I would not be surprised if Apple saw these trends and realized that the upgrade cycle was too long to justify continuing to offer it with Intel (current or updated). Apple Silicon will be so much faster than Intel at the tasks so many current iMac Pro owners use that an AS iMac Pro would be what gets them to replace their 2017 machines, not a slightly faster Intel model.

 

[dgdosen]

Could this just be Apple leaking noise in an attempt to rain on Intel's Alder Lake Mobile parade?

 

[leman]

Could this just be Apple leaking noise in an attempt to rain on Intel's Alder Lake Mobile parade?

I don’t think Apple cares. The markets are sufficiently different.

 

[deconstruct60]

Most people I know who purchased the iMac Pro were software developers, photo editors and podcasters who purchased it for the power, the quiet and because it had a 5K display. ,,,,

I would not be surprised if Apple saw these trends and realized that the upgrade cycle was too long to justify continuing to offer it with Intel (current or updated).

Doubtful. The bigger problem for the iMac Pro was that it wasn't exactly the same 'corner' they painted themselves into with the Mac Pro 2013 , but it had a similar boundaries ( a 'corner' that was bigger room to move around in but still a corner. )

CPU wise they had issues. The W-2100 was followed by W-2200 ( which basically same thing clocked higher at a larger TDP]. Larger TDP doesn't help with iMac Pro because at threshold where trying to keep it quiet with a air exhaust vent limited by how much can hide it behind the pedestal arm. Going to W-2200 would hae allowed Apple to lower the price (or keep price constant and shift higher core counts to lower prices). If the long term Apple Silicon intend was to keep the "upscale iMac big screen" just as high then lowering it would be at odds with their objectives also. [ If going to a dual package they are not shooting for lower price points. ]

Xeon W-3300 ( ice lake) is even more off the scale TDP problem. ( Intel isn't even attempting a W-2300 solution ) . And it was very late.


GPU wise they had upgrade issues. RNDA 1 didn't really include upscale GPUs. There were some things a 5700 was incrementally better and other things it was not. Clocking the Vega higher would run into TDP issues ( which was why it was downclocked in the first place). The Vega II ( Vega20 ) versus Vega64 ( Vega10) wouldn't work well if in process of turning off Vega II Pro cards for 6800/6900 ones. Footprint and expense would be bigger with two HBM stacks.


If Apple could have upgraded the iMac Pro in 2020 it would have made sense. ( at around the same 2H 2020 the regular iMac did. They didn't. Enough overlap with a 10 core Gen 10 CPU and 5700 HBM pulled back to just the regular iMac. Intel and AMD had parts... just not parts that would 'top' the next more mainstream components. )

Apple could have done an incremental. The did "cheap" incremental moves by raising the max RAM and SSD capacity. If Intel's and AMD's roadmaps in 2017 had been right and the W-2200 has been a performance uplift with little to no TDP penalty and AMD rolled out a full line up. I wouldn't be surprised is Apple did the intial ground work to do a iMac Pro upgrade and then quit when the parts turned into "overclocked, barn burners". So impact like the MP 2013 of upgrade inaction because couldn't get parts wished they could get from "early roadmaps".

Apple Silicon will be so much faster than Intel at the tasks so many current iMac Pro owners use that an AS iMac Pro would be what gets them to replace their 2017 machines, not a slightly faster Intel model.

After 5 years you'd hope something built new in 2022 was faster. When the Gen 12 based (Alder Lake ) Xeon W-14xx arrive later in 2022 , then they will probably whip that old W-2100 too. Or a Zen4/Zen3 AMD alternative.
If Apple stumbles around until late Fall 2022 to get the big screen iMac out with just a M1 Pro and Mas it isn't like they'll be kicking sand in the face of the contemporary stuff on the other side in the desktop space.

 

[deconstruct60]

Could this just be Apple leaking noise in an attempt to rain on Intel's Alder Lake Mobile parade?

Since Apple claims they have "desktop" performance .... more like trying to keep up with Alder Lake Desktop.
According to Apple not suppose to be comparing to Intel's mobile offerings. Apple stuff is suppose to be better than a desktop processor from those "other guys".

 

[deconstruct60]

Die yields have everything to do with this. Yes, chipmakers can and do bin down based on competition and price. But why not sell an 18-core at a higher price instead?

Why didn't the Mac Pro 2013 have 24 cores inside? Because Apple had other design objectives for the overall chassis. One look at the iMac 24" and it is a very easy path to Apple putting some self imposed caps on a iMac 27' that means pragmatically better to stop at 12 cores.

Just like the report leaving off the GPU core count ( which would relatively obvious indicate whether new baseline die or just a dual ) , there is little about he chassis they are trying to squeeze this into in the report.

Apple cuts iMac short on thermals would not be a shocker. Should they do it? No. Will they do it? Maybe (They have done it before. So there is a track record. )

There is zip , nothing , nada in this "iMac only" report that stops Apple from making 16-20 variants of a two chiplet/tile package. There is also nothing that says those "have to" fit inside an iMac.

And if barely going to able to produce enough 16 , 18 , 20 cores for an "Mac Cube" then way strain it further with "iMac Max" demand? Without going to two Max sized dies per package, Apple is coming up short for the laptops. Putting them in iMacs isn't going to increase the supply. If also do Minis Doing later intros to bigger than 12 could be an option if just want to get the bulk of the iMacs out the door with Pros and plain Maxes.

The longer Apple squats on that last Intel iMac. The longer they have to support it into the future.

Binning down to 12 cores from a total of 20? Why? Why not 18 cores? Why not 16? Why not 14? Why bin down 40% of the CPU cores when 5nm yields are so good?

1. It will be more affordable to end users . At a $100/core mark up that would be $600-800 difference in price ( on top of something that is already in the > $1,000 range .
[ There is about zero chance they are loosing money at whatever price they set to 12 core model do. and they print even higher margins with those $100/core mark-ups. ]

2. You got to get a wafer start to get a yield. AMD is ramping on 5nm in the Fall. Nvidia is ramping on 5nm in the Fall. There are number of players ramping on 5nm in the Fall. iPhone SE 5G . The other Macs that may not move off off a non 5nm M-series . etc. If launching on a product with high overlap that is going to consume most of the projected two-die packages ( Mac "cube" ) then getting 12 cores could be enough to soak up volume if the other product unexpectedly comes up short, but limited enough not to further exacerbate the problem if both go over expectations.

The premise that Apple has infinite 5nm wafer volume available is suspect. If 12's fit the chassis then make them and put them in.

Unless Apple plans to introduce 20, 18, 16, and 14 core versions, it makes no sense to deliberately disable 40% of good cores.

There is not absolute requirement that all of 14 , 16 , 20's go into iMac Pros?

If there are any quad packages then will be sucking up those Max sized dies (and therefore wafers ) at 4x the pace they M1 Mac does.

The 40% of good cores is completely out of context. That isn't a cost killer in the context of the space that the GPU cores soak up on a Max sized die. The P cores ( sans L2) take up about as much die space as the Thunderbolt complex in the upper right.

Apple Announces M1 Pro & M1 Max: Giant New Arm SoCs with All-Out Performance

www.anandtech.com

And in that picture only really looking at about half of the complete Max die. As an overall percentage of the die this is not a huge percentage of overall die space used. So not a huge hit to cost (or cost recovery).

If turn off those 4 cores at the bottom near the GPU complex then have a thermal buffer between the two major sections. If trying to run under a tighter than initially designed for thermal window that would help.

 

[CWallace]

Rene Ritchie weighed in and he believes this SoC will be an M1 Max with 12 CPU cores (10 P and 2 E) and 32 GPU cores with 64GB of RAM and possibly more. He does not believe this is a binned-down version of the "M1 Max Duo".

 

[mr_roboto]

Rene Ritchie weighed in and he believes this SoC will be an M1 Max with 12 CPU cores (10 P and 2 E) and 32 GPU cores with 64GB of RAM and possibly more. He does not believe this is a binned-down version of the "M1 Max Duo".

10P+2E seems unlikely, IMO. So far, M1 P cores have always been clusters of four sharing a single L2 cache. While Apple did use a 2-core P cluster in A14, 4+4+2 is a weird configuration with some performance asymmetry between P cores (because two of them would have more L2 cache per core than the rest).

My guess (which I've posted about elsewhere, see @cmaier 's sig) is that if it's a chip designed only for desktops, they may have just deleted the E cluster and replaced it with another 4-core P cluster, giving us 12 P cores.

All this assumes the rumor is correct, which it may not be.

 

[cmaier]

Rene Ritchie weighed in and he believes this SoC will be an M1 Max with 12 CPU cores (10 P and 2 E) and 32 GPU cores with 64GB of RAM and possibly more. He does not believe this is a binned-down version of the "M1 Max Duo".

Let’s all play “floor planner” and come up with die layouts.

 

[deconstruct60]

Rene Ritchie weighed in and he believes this SoC will be an M1 Max with 12 CPU cores (10 P and 2 E) and 32 GPU cores with 64GB of RAM and possibly more. He does not believe this is a binned-down version of the "M1 Max Duo".

that doesn't make sense. The current P cores are in two 4 core clusters. To split. 10 into equal halves would be 5 each. Apple has done odd=number three cores before, but that seems a bit quirky. Plus pretty good chance oversubscribing the shared L2 cache. The other bandwidth issue have is that P clusters were throttled to keep bandwidth available for the 32 GPU + NPU + ProRes bandwidth throughput. Where is the extra bandwidth coming from for the additional cores?


If they were off on a variant with same basic parts but bigger and reflowed die then. 12P and 2E would make more sense. 2P 'spares' wouldn't be high overhead ( with 32 GPU cores 4P spares really isn't either) and "cheaper" monolithic die package. Three clusters of 4 and at least doesn't oversubscribe the L2 . Could probably get by if don't "walk and chew gum" at the same time at high concurrency of function units on the backhaul memory bus.

The core problem though if creating 2-die packages anyway for the Mac Pro what did you actually save here????
Have yet one more die to do and possibly fewer Mac products to put it to get to high volume and economies of scale.
[ Is it heading for the larger Mini also? It would have to be something else with very low internal expansion. Macbook Max 16" ? ]

If this was some. M2 Max where went from N5 to N4 and picked up some minor density uplift and was doing reflows anyway and used in a wider scale product refresh, then fine. M1 though... that's odd.

 

[deconstruct60]

Let’s all play “floor planner” and come up with die layouts.

Probably easier to just put it on the 'other side' of the 'bottom half' GPU cluster and re-arrange the NPU+Prores+video section somewhat narrower and extended the die out the bottom more. square off the die P core cluster and perhaps some more general purpose I/O that might be useful.


If there is some CPU L2 to L2 latency hiccup then evict the upper NPU+Prores+video to the bottom and tightly and re-arrange those two clusters together to square off new longer bottom.

 

[senttoschool]

10P+2E seems unlikely, IMO. So far, M1 P cores have always been clusters of four sharing a single L2 cache. While Apple did use a 2-core P cluster in A14, 4+4+2 is a weird configuration with some performance asymmetry between P cores (because two of them would have more L2 cache per core than the rest).

My guess (which I've posted about elsewhere, see @cmaier 's sig) is that if it's a chip designed only for desktops, they may have just deleted the E cluster and replaced it with another 4-core P cluster, giving us 12 P cores.

All this assumes the rumor is correct, which it may not be.

Not sure if it makes sense to delete the E clusters. The E clusters are still super useful, efficiency-wise, for background OS tasks on desktop. Many of these tasks don't need P cores at all.

But it'll be a hell of a chip if it's 12P cores.

 

[deconstruct60]

My guess (which I've posted about elsewhere, see @cmaier 's sig) is that if it's a chip designed only for desktops, they may have just deleted the E cluster and replaced it with another 4-core P cluster, giving us 12 P cores.

All this assumes the rumor is correct, which it may not be.

4 E cores take up about the same space at one P core. Deleting 2 E cores doesn't save much of anything significant of die space (or causing some significant cost reduction). The second NPU+ProRes+Video cluster is using far more space. Every 12 core user is going to use double ProRes? Probably not and yet it is there on the die.

If start up activity monitor and scroll through the long list of instantiated processes .... can you actually get through a whole 24 period with the Mac up and working without any of those ever hitting a E core? The E cores are used on a daily basis on all M-series macs. It isn't sexy tech porn benchmark work , but it is daily work.

 

[senttoschool]

Why didn't the Mac Pro 2013 have 24 cores inside? Because Apple had other design objectives for the overall chassis. One look at the iMac 24" and it is a very easy path to Apple putting some self imposed caps on a iMac 27' that means pragmatically better to stop at 12 cores.

Just like the report leaving off the GPU core count ( which would relatively obvious indicate whether new baseline die or just a dual ) , there is little about he chassis they are trying to squeeze this into in the report.

Apple cuts iMac short on thermals would not be a shocker. Should they do it? No. Will they do it? Maybe (They have done it before. So there is a track record. )

There is zip , nothing , nada in this "iMac only" report that stops Apple from making 16-20 variants of a two chiplet/tile package. There is also nothing that says those "have to" fit inside an iMac.

And if barely going to able to produce enough 16 , 18 , 20 cores for an "Mac Cube" then way strain it further with "iMac Max" demand? Without going to two Max sized dies per package, Apple is coming up short for the laptops. Putting them in iMacs isn't going to increase the supply. If also do Minis Doing later intros to bigger than 12 could be an option if just want to get the bulk of the iMacs out the door with Pros and plain Maxes.

The longer Apple squats on that last Intel iMac. The longer they have to support it into the future.





1. It will be more affordable to end users . At a $100/core mark up that would be $600-800 difference in price ( on top of something that is already in the > $1,000 range .
[ There is about zero chance they are loosing money at whatever price they set to 12 core model do. and they print even higher margins with those $100/core mark-ups. ]

2. You got to get a wafer start to get a yield. AMD is ramping on 5nm in the Fall. Nvidia is ramping on 5nm in the Fall. There are number of players ramping on 5nm in the Fall. iPhone SE 5G . The other Macs that may not move off off a non 5nm M-series . etc. If launching on a product with high overlap that is going to consume most of the projected two-die packages ( Mac "cube" ) then getting 12 cores could be enough to soak up volume if the other product unexpectedly comes up short, but limited enough not to further exacerbate the problem if both go over expectations.

The premise that Apple has infinite 5nm wafer volume available is suspect. If 12's fit the chassis then make them and put them in.




There is not absolute requirement that all of 14 , 16 , 20's go into iMac Pros?

If there are any quad packages then will be sucking up those Max sized dies (and therefore wafers ) at 4x the pace they M1 Mac does.

The 40% of good cores is completely out of context. That isn't a cost killer in the context of the space that the GPU cores soak up on a Max sized die. The P cores ( sans L2) take up about as much die space as the Thunderbolt complex in the upper right.

Apple Announces M1 Pro & M1 Max: Giant New Arm SoCs with All-Out Performance

www.anandtech.com

And in that picture only really looking at about half of the complete Max die. As an overall percentage of the die this is not a huge percentage of overall die space used. So not a huge hit to cost (or cost recovery).

If turn off those 4 cores at the bottom near the GPU complex then have a thermal buffer between the two major sections. If trying to run under a tighter than initially designed for thermal window that would help.

I'm not convinced.

Again, it makes no sense to glue two M1 Max together and only enable 12-cores (with no plans to enable 14,6,18,20 bins). TSMC's 5nm yields should be plenty good to enable higher core counts.

 

[Malus120]

I'm not convinced.

Again, it makes no sense to glue two M1 Max together and only enable 12-cores (with no plans to enable 14,6,18,20 bins). TSMC's 5nm yields should be plenty good to enable higher core counts.

I don't think it makes sense to be convinced of anything at this point. I see far too much of what I hope is just good natured discussion, but at times is beginning to look a bit like bickering over what are, at the end of the day, just rumors that didn't even make it to the front page (not trying to cast doubt on the specific credibility of any of these individuals, it just is what it is).

If someone had told me in September that:
1. The M1 Max would have 400Gbps of unified memory bandwidth
2. The M1 Pro and Max would have 200Gbps of that available to the CPU (!)
3. The M1 Pro/Max would feature a 6/2 and 8/2 split of P to E cores (instead of 6/4, 8/4 or some other combination that used the existing 4/4 layout of P to E cores.)
4. The E cores would run at twice the frequency when fully loaded so as to complete background tasks at the same speed as the original M1.

I don't think I would have believed them as very little of that lined up with what we saw and could infer from the OG M1 design.

All of which is to say that while there's nothing wrong with a little healthy speculation, I don't think anyone who is not under NDA really knows what Apple can and can't do with Apple Silicon / whether the building blocks we see in the M1/Pro/Max are really all that Apple has. It's entirely possible that, much like how the M1 Pro/Max built on the M1 design while also breaking with it in several fundamental ways, Apple may use the iMac as an opportunity to begin the process of retooling Apple Silicon for desktops, using the M1 Pro/Max as a base but changing aspects of the design such as the ratio of P/E/GPU cores. It's also entirely possible that they just want to reuse the M1 Pro/Max, whether that means using it as is, doubling the core counts, or something in-between.

We just have to wait and see (and speculate)

 

[crazy dave]

I don't think it makes sense to be convinced of anything at this point. I see far too much of what I hope is just good natured discussion, but at times is beginning to look a bit like bickering over what are, at the end of the day, just rumors that didn't even make it to the front page (not trying to cast doubt on the specific credibility of any of these individuals, it just is what it is).

If someone had told me in September that:
1. The M1 Max would have 400Gbps of unified memory bandwidth
2. The M1 Pro and Max would have 200Gbps of that available to the CPU (!)
3. The M1 Pro/Max would feature a 6/2 and 8/2 split of P to E cores (instead of 6/4, 8/4 or some other combination that used the existing 4/4 layout of P to E cores.)
4. The E cores would run at twice the frequency when fully loaded so as to complete background tasks at the same speed as the original M1.

I don't think I would have believed them as very little of that lined up with what we saw and could infer from the OG M1 design.

All of which is to say that while there's nothing wrong with a little healthy speculation, I don't think anyone who is not under NDA really knows what Apple can and can't do with Apple Silicon / whether the building blocks we see in the M1/Pro/Max are really all that Apple has. It's entirely possible that, much like how the M1 Pro/Max built on the M1 design while also breaking with it in several fundamental ways, Apple may use the iMac as an opportunity to begin the process of retooling Apple Silicon for desktops, using the M1 Pro/Max as a base but changing aspects of the design such as the ratio of P/E/GPU cores. It's also entirely possible that they just want to reuse the M1 Pro/Max, whether that means using it as is, doubling the core counts, or something in-between.

We just have to wait and see (and speculate)

Agreed, hopefully everyone is just having fun here? Obviously we all have somewhat more information now that the Max/Pro have released and there are those who have more experience than the rest of us, so the speculation now can be more informed in general (and in particular). But all the debating about what logically makes the most sense going forward is still just rank speculation.

I mean we don’t actually know the significance or relevance of the 12 cores code snippet, even if accurate we don’t know the associated GPU core count, we don’t know if that represents the base of a new die (ie a cutdown die) or set of configurations which may or may not be destined for the 27” iMac, we don’t know how the 27” is going to be redesigned or it’s thermal capacity, etc … there are waaay too many known unknowns that there don’t even have to be unknown unknowns! (And there may be those too)

Truthfully this is one of the reasons why I described myself as impatient in another thread - it isn’t because I personally want to buy all of these future machines.?? I’m just really interested to see what Apple does here, to see how they think now that they have control over their own processors.

 

[iPadified]

Who is this iMac for? Video editors? Ordinary office people needing a larger screen than 24 inch? Scientists/engineers needing compute? 3D modellers and rendering? The target audience is critical regarding how many cores of this and that that will be added. 12-core CPU does not seem to be a huge step up from M1 Max in many application areas.